CogITo Smart Journal

Predictive Maintenance of Heavy Equipment Machines using Neural Network Based on Operational Data

2024-12-04T08:13:16+00:00

Preventive maintenance is a routine maintenance strategy that aims to maximize equipment life cycle and prevent unplanned downtime which causes increased repair costs. When carrying out this maintenance, error in selecting machines need to be anticipated to avoid company losses. This research aims to reduce human error in machine selection for preventive maintenance using deep learning. The dataset used in this research is operational data of heavy equipment machine dataset from one of the palm oil companies in Indonesia with 9 independent features and 1 dependent feature. Dependent feature is a target feature contain two target classes representing effective and ineffective machines. The dataset in this study contains outlier, feature scales that are very different, and imbalanced data class. To handle outlier and standardise data scale, the Z-score method is used. Meanwhile, the over sampling method is used to handle imbalanced data classes. To obtain the best model performance, the number of epochs and two types of optimizers (adam&adamax) of neural network are selected. In selecting the number of epochs, experiments were carried out using 100 epochs. This research obtained the linearity relationship between the number of epochs and accuracy with the accuracy values using Adam and Adamax optimizers were 94.82% and 93.11% at the 100th epoch.

Recapitulation of Lecturer’s Attendance Using Android-Based Fingerprint At Dipa Makassar University

2025-06-19T14:50:20+00:00

An academic information system is expected to provide information to lecturers, students, and administrators. There are several parts included in the academic information system, including lecturer teaching schedules, lecturer attendance systems (monitoring), and student lecture schedules. These three things are related to the results of the recapitulation of lecturer attendance for reporting on the academic section institution. Dipa Makassar University, the process of recapitulating lecturer attendance still uses a manual system, namely, officers record lecturer attendance at each session and then input it into the Excel application. This process is inefficient because the work is repetitive and time-consuming. When lecturers are delayed, there is no information provided to students, causing them to wait for long periods. In this study, the researcher developed an Android-based fingerprint application for recapitulating lecturer attendance and tardiness. The testing method applied to evaluate the application is the Black Box testing method. Based on the overall testing results, it can be concluded that the application functions properly in accordance with the system requirements and is free from errors. The testing focused on the application's input, process, and output features without examining the internal structure of the program. All tested features produced valid results and met the expected outcomes. With the presence of this application, both lecturers and students can access class schedules at any time through the Android platform. Additionally, the application provides notifications to students in case the lecturer arrives late, which significantly improves communication between lecturers and students.

Comparative Analysis Of Convolutional Neural Network Models For Digital Image-Based Melanoma Classification

2025-06-19T14:32:30+00:00

Melanoma is one of the most malignant forms of skin cancer, with an incidence rate of 7.9% in Indonesia. Traditional biopsy-based diagnosis, though crucial, is invasive and time-consuming, creating barriers for early detection. To address this issue, this research compares two Convolutional Neural Network (CNN) models for digital image-based melanoma classification. The study utilized a publicly available dataset from Kaggle, consisting of 17,805 images (melanoma and non-melanoma), which were divided into training, validation, and testing subsets. The models were trained using the Adamax and SGD optimizers for 100 epochs. The performance of the models was evaluated based on accuracy, loss, precision, recall, and F1-score. The CNN model with the best architecture, which consisted of two fully connected layers, achieved an accuracy of 93.18% and a loss of 0.1636, outperforming the alternative model. These results confirm the effectiveness of CNN models in classifying melanoma images and support the development of a web-based platform that allows users to upload or capture images for rapid and non-invasive detection.

Assessing Information Security Readiness in Indonesian Fintech Companies Using KAMI Index 5.0 Framework

2025-06-19T14:33:36+00:00

The development of Indonesian financial technology (fintech) has transformed the financial industry paradigm but has also introduced significant information security risks, particularly for technology-based companies. The fintech companies should establish IT governance through an Information Security Management System (ISMS) which adheres to international standards, ensuring the confidentiality, integrity, and availability of information. This work adopts a qualitative approach deploying observations, interviews, and literature reviews on Indonesian fintech companies, especially digital banking fields, payment gateways, and digital wallet platforms. This study is to identify information security risks and assess the readiness and feasibility of implementing ISO/IEC 27001:2022 using the KAMI Index 5.0, which evaluates domains such as policy, governance, risk management, access control, incident management, asset management, and personal data protection. The research findings indicate that the electronic system of fintech companies plays a strategic role in supporting sustainability and business growth, with an implementation score of 809 and a fairly good level of information security feasibility. In conclusion, this reflects the company’s readiness for further information security implementation. The system not only supports basic operations but also serves as a key element in achieving business objectives, both internally and externally, including regulators, banking partners, and customers.

Optimizing Network Traffic Classification Models with a Hybrid Approach for Large-Scale Data

2025-09-02T02:11:12+00:00

The escalating threat of cyberattacks necessitates the development of intrusion detection
models that are both accurate and computationally efficient for large-scale network traffic. To
address this issue, this study proposes a hybrid approach combining Autoencoder, Convolutional
Neural Network (CNN), and XGBoost as an adaptive and lightweight solution for network traffic
classification. The key contribution of this research lies in the design of a multi-stage pipeline
that performs dimensionality reduction, feature extraction, and final classification. The model
was evaluated using the Moore Dataset, which contains complex and high-dimensional network
traffic data. The experimental results indicate that the proposed hybrid model achieved a
classification accuracy of 99.20% with a testing time of only 0.09 seconds. Furthermore, the
pipeline significantly reduced computational load compared to single CNN or XGBoost models.
These findings demonstrate that the hybrid approach not only offers high classification
performance but also enhances scalability and efficiency, making it suitable for real-world
implementation in modern network security systems. Overall, the proposed model presents a
promising and practical solution for advancing future intrusion detection systems.

Forecasting the Highest and Lowest Prices in Financial Markets Using a VMD-LSTM Hybrid Model

2025-08-04T01:13:26+00:00

Accurate forecasting of the lowest and highest prices in financial markets poses a considerable challenge due to the inherent nonlinear behaviour, non-stationarity, and high noise levels of financial time series data. Most prior studies focus only on closing prices, with limited attention to the simultaneous prediction of high and low prices. Yet, predicting the lowest and highest prices is essential for investors to make informed trading decisions. To address this gap, this study proposes a hybrid DL framework that integrates VMD and LSTM networks for predicting daily high and low prices simultaneously. This study used 12 years of daily data from three diverse assets: AUD/USD, TLKM, and XAU/USD. The data underwent preprocessing, VMD-based decomposition, and were input into the LSTM. The dataset was split 80% for training and 20% for testing. Experiments varied the number of decomposition modes (K = 7, 10, 12) and sliding window sizes (5, 15, 30, 45, 60, 90). Results show that the VMD-LSTM model exhibits improved performance in most of the tested scenarios compared to traditional LSTM. These findings underscore that the use of VMD decomposition can help enhance the accuracy of forecasting the highest and lowest prices in the financial market.

Smart Assistive Stick with Arduino and Multidirectional Ultrasonic Sensors for Intelligent Obstacle Detection and Navigation

2025-09-02T03:49:13+00:00

Blindness or visual impairment restricts spatial awareness and increases the risk of collisions, falls, and mobility challenges. This study presents the design and development of a Smart Assistive Stick with Arduino and multidirectional ultrasonic sensors for intelligent obstacle detection and navigation. Unlike conventional white canes that provide only short-range tactile feedback, the proposed system employs multidirectional sensing to detect obstacles from various directions within a range of 0.1 to 4 meters. Intelligent feedback is delivered through both haptic and auditory signals, with an average response delay of only 200 ms, ensuring timely and reliable navigation assistance. Testing showed detection accuracy exceeding 85%, continuous battery life of 6–8 hours, and a total device weight of 600 grams, making it lightweight and suitable for daily use. While performance decreases in noisy environments due to ultrasonic interference, the system demonstrates novelty in extending detection range, incorporating multidirectional sensing, and providing intelligent real-time feedback. These contributions establish the smart assistive stick as a more effective and user-friendly mobility aid compared to traditional solutions.

Transparency and Trust in Minahasa Tourism Advertising using Blockchain

2025-10-09T05:23:01+00:00

Abstract

Tourism plays a vital role in driving economic growth, and Minahasa holds strong potential to optimize this sector. However, challenges remain in digital advertising, particularly regarding transparency and consumer trust. This study investigates the impact of blockchain technology on transparency, trust, and the effectiveness of digital advertising in Minahasa’s tourism industry. A quantitative explanatory design was employed using Partial Least Squares Structural Equation Modeling (SEM-PLS), with data collected from 150–250 respondents through purposive and snowball sampling techniques.The findings reveal that blockchain significantly influences all key variables. It enhances advertising transparency (T-statistic = 36.738, p = 0.000), strengthens consumer trust (T-statistic = 33.164, p = 0.000), and improves advertising effectiveness (T-statistic = 28.400, p = 0.000). These results highlight blockchain’s capacity to provide immutable records, ensure data authenticity, and optimize ad performance through verifiable real-time information. This study confirms that blockchain can serve as a strategic tool to promote transparent, trustworthy, and effective digital advertising in tourism. The findings provide practical insights for tourism stakeholders and contribute to academic discussions on technology-driven marketing innovation.

Funnel-Based Predictive Modeling for Forecasting Student Admissions in Higher Education

2025-10-28T05:13:58+00:00

Forecasting student admissions remains a challenge due to fluctuating online engagement and complex administrative processes. Existing predictive models rarely integrate website behavioral data with institutional admission funnels, resulting in lower accuracy. This study bridges that gap by combining web analytics from Google Analytics 4 (GA4) with administrative enrollment funnel data from the admission of new students (Penerimaan Mahasiswa Baru/PMB) system to develop a unified predictive framework. The approach strengthens forecasting by aligning digital behavior with verified enrollment milestones. A quantitative explanatory design was employed, applying Pearson correlation to identify linear relationships and Seasonal ARIMA (SARIMA) to model cyclical admission trends. The dataset includes GA4 metrics sessions, engagement rate, bounce rate, and events per session and PMB funnel stages from account creation to confirmed enrollment. Results reveal strong correlations (r > 0.9, p < 0.001) between digital engagement and mid-funnel conversions, while SARIMA achieved its highest accuracy for early-stage predictions (MAPE ≈ 19%). Forecasts for final outcomes were less accurate, reflecting administrative variability. These findings confirm that web engagement metrics are reliable leading indicators of student interest and mid-stage commitment. This research establishes a reproducible pipeline unifying web analytics (GA4) with institutional funnel data (PMB), providing empirical evidence that digital engagement is a reliable leading indicator of early and mid-stage commitment, thereby forming a novel and adaptable foundation for data-driven enrollment planning.

Examining Lecturers’ Learning Management System Usage Using TAM: Eastern Indonesia Case Study

2025-11-27T04:01:44+00:00

The implementation of Learning Management Systems (LMS) in higher education institutions continues to increase in line with the growing demand for flexible digital learning, with the assumption that LMS is an easy-to-use platform that will be naturally accepted by lecturers. This study aims to analyze the factors that influence the adoption of LMS among lecturers at higher education institutions in Eastern Indonesia. This study uses a quantitative cross-sectional survey. The research instrument, comprising 25 items classified into five constructs —Constructivist Pedagogical Beliefs, Traditional Pedagogical Beliefs, Perceived Ease of Use, Perceived Usefulness, and LMS Use —was administered to lecturers at a private university in North Sulawesi. Using the Partial Least Squares-Structural Equation Modeling approach, this study incorporates the Technology Acceptance Model with a constructivist and traditional pedagogical belief orientation. The results show that three of the eight variables significantly influence LMS usage. The findings indicate that constructivist pedagogical beliefs and perceived usefulness have a significant influence on LMS adoption, whereas traditional pedagogical beliefs do not have a significant impact. These results have practical implications for universities in designing training policies and strategies to optimize LMS usage.

Design and Implementation of Full-Stack Conference System for Streamlined Administrative Workflows

2025-11-13T14:19:34+00:00

This study presents the design, implementation, and evaluation of the 11ISC Conference Management System (CMS), a full-stack web application developed to address the fragmented administrative workflows of the 11th International Scholars Conference. Using the Design Science Research methodology, the system was created in response to recurring challenges such as manual registration, accommodation and transportation coordination, and the time-intensive preparation of Letters of Acceptance. The CMS was evaluated through blackbox functional testing covering twelve primary use cases, all of which passed successfully, including participant registration, payment verification, automated LoA generation, QR-based check-in, and accommodation assignment. Administrator feedback indicated substantial process improvements, with the automated LoA module reducing preparation time by up to 90 percent and integrated room and check-in management significantly decreasing errors associated with the previous spreadsheet-based workflow. Deployed during the conference, the system supported more than 220 participants and over 180 paper submissions, providing real-time dashboards and unified data management. The results demonstrate that the CMS enhances efficiency, accuracy, and coordination, offering a practical and replicable solution for academic event management in similar institutional contexts.

Benchmarking Five Machine Learning Models for Accurate Steel Plate Defect Detection

2025-06-19T14:41:42+00:00

Early detection of defects in steel plates is essential to ensure structural integrity and product quality in the metal manufacturing industry. This study compares the performance of five machine learning algorithms Support Vector Classifier (SVC), Nu-Support Vector Classifier (NuSVC), Decision Tree (DT), Random Forest (RF), and CatBoost (CB) to classify seven categories of steel plate defects using 26 technical features from a publicly available dataset on Kaggle. The preprocessing pipeline included outlier detection (IQR method), class imbalance correction using SMOTE, and feature normalization via StandardScaler. The models were evaluated using classification metrics such as Accuracy, Precision, Recall, F1-Score, ROC-AUC, and Log Loss. Results revealed that the CatBoost algorithm achieved the most balanced and consistent performance, with an AUC of 0.93, accuracy of 68.3%, and the lowest Log Loss value (0.786). In contrast, the Decision Tree showed severe overfitting with perfect training performance but poor generalization (Log Loss = 15.72). This study highlights the promise of CatBoost as an interpretable and efficient solution for automated defect detection in steel manufacturing, while also offering transparent reproducibility pathways for further research.

Predictive Linear Regression Model for Premature Birth Risk Assessment System

2025-10-28T06:42:25+00:00

Preterm birth is a major cause of neonatal mortality in Indonesia and is influenced by multiple maternal factors. Early prediction models are crucial for supporting timely clinical decision-making and reducing adverse maternal–infant outcomes. The method of this study developed a linear regression–based predictive model using 915 pregnancy medical records from Dr. H. M. Ansari Saleh Regional Hospital, Banjarmasin (2020–2022). The workflow included data preprocessing, feature selection, Min-Max normalization, and experimentation with various train–test split ratios (90:10 to 50:50). Model performance was evaluated using R², Adjusted R², MAE, MSE, RMSE, and MAPE metrics. As the results, the 70:30 split ratio achieved the best accuracy of 89.05% and AUC of 98.10%, with low prediction errors. Optimizations with Adamax and Nadam enhanced stability and reduced MAPE to 1.95%. The optimized linear regression model reliably predicts preterm birth risk and is suitable for clinical decision support, particularly in resource-limited settings.

Text Similarity Analysis for Evaluating Alignment Between Lesson Plans and Teaching Reports

2025-12-22T08:19:24+00:00

RPS (Rencana Pembelajaran Semester, or called Lesson Plans) is a class activity planning document in the higher education learning process that includes learning outcomes, methods, learning strategy, and evaluation criteria. It is created by the lecturers in charge of the course and coordinated with the relevant department. This document needs to be monitored throughout the semester for its conformity with the implementation document (Borang Pelaksanaan Perkuliahan (BPP)). It was done manually through our eRPS system, but it requires a lot of effort and precision and is not time-efficient. This research focused on evaluating the effectiveness of several content-based text similarity methods to detect RPS conformity compared with the BPP, or called Teaching Reports document. The Boyer-Moore (B), Rabin-Karp (R), Jaccard (JC), Jaro-Winkler (JW), Smith-Waterman (SW), Knuth-Morris-Pratt (K), Levenehtein cosine similarity (C), Dice (D), Jaro (J), and Soundex (S) algorithms were evaluated in this paper. In the vector-based similarity method, TF-IDF was used. The evaluation of 11 string-matching algorithms across four scenarios demonstrated clear performance trends. Fuzzy algorithms (SW with accuracy 0,845–0,870, and JW with accuracy 0,840-0,850) achieved the highest accuracy in a single row of lecturer scenario, while exact/pattern-based algorithms (B, K, and S with accuracy 0,8625–0,8725) on a combination of all rows of lectures with minimal variance (≈0,005–0,015). Pre-processing benefits fuzzy algorithms (+2.5%) but is neutral for exact/pattern-based algorithms. The combined scenario improves the exact/phonetic algorithms (+6–7%) but reduces the fuzzy performance algorithm (−10–14%). The optimal thresholds were generally 40–50%, except for JW and J, which were 65%.

Neural Dynamic Network for Brain Tumor Classification: An Attention-Based Feature Selection Approach

2025-11-11T13:20:43+00:00

Magnetic Resonance Imaging (MRI) plays a vital role in the early detection of brain tumors. However, standard Convolutional Neural Network (CNN) models often struggle to extract truly relevant features from complex MRI structures. This limitation creates a gap in achieving robust and clinically interpretable classifications, as feature redundancy and weak attention toward tumor-specific regions may reduce diagnostic reliability. To address this gap, this study introduces a Neural Dynamic Network (NDN) that integrates EfficientNetV2S with a dynamic attention-based mechanism to adaptively highlight informative features while suppressing noise. The proposed model was evaluated using a 5-fold cross-validation scheme and tested on unseen data. Compared with the baseline CNN, the NDN consistently demonstrated higher accuracy, precision, recall, and F1-score across folds and final testing, reflecting improved robustness and balanced sensitivity. NDN yielded significant improvements, with the 5-fold validation averaging an accuracy of 88.44%, a precision of 87.84%, a recall of 87.88%, and an F1-score of 87.82%. Beyond numerical performance, interpretability analysis utilizing Grad-CAM demonstrated that NDN generates more concentrated and clinically consistent heatmaps. In contrast, the baseline CNN produced dispersed activations that exhibited less alignment with tumor regions. Overall, the findings confirm that incorporating a dynamic attention-based mechanism substantially enhances both feature selection and visual interpretability. This makes the NDN architecture more reliable for MRI-based brain tumor classification and highly suitable as a decision-support tool in clinical workflows.

Hyperparameter Tuning Exploration to Maximize MobileNet Performance in Classification Kidney Tumor

2025-11-10T15:07:38+00:00

The main focus in this research is how to develop the MobileNet architecture in order to produce a kidney tumor classification model that is accurate, resistant to overfitting, and remains consistent with variations in datasets and training parameters. This study aims to develop MobileNet architecture to produce a software model that can precisely identify with high accuracy, perform kidney tumor classification, and avoid failure in generalizing new data called overfitting, as well as evaluate the difference in accuracy generated from several variations of datasets and parameters. The method used in this study is MobileNet with hyperparameter tuning and fine-tuning, and it was compared with the MobileNet Baseline method. The dataset consists of 12,446 images classified as Normal, Cyst, Stone, and Tumor, collected from Kaggle. The findings of this study on the division of the 80:10:10 ratio of the proposed method image data resulted in 100% accuracy, 100% precision, 100% recall, and 100% F1-Score. This study is expected to produce architecture modifications that can classify kidney tumors with high accuracy so that the hypothesis is achieved. In addition, various approaches in medical image analysis using deep learning have shown better results in identifying various tumors, especially this research in the classification and detection of kidney tumors.

Forecasting Medium Rice’s Retail Price with Machine Learning in Gorontalo Province

2025-11-05T08:18:07+00:00

The stability of rice prices is essential for food security in Indonesia, particularly in Gorontalo Province where volatility has increased in recent years. This study develops a machine learning-based forecasting framework using Decision Tree, Random Forest, and K-Nearest Neighbors (KNN) to estimate next-day retail prices. A harvest-season indicator was incorporated to capture agricultural seasonal patterns. Data preprocessing included feature engineering, data cleaning, exploratory analysis, and chronological splitting to maintain temporal order. Model performance was assessed using RMSE and MAPE. The optimized KNN model achieved the highest accuracy, with an RMSE of 96.76 and a MAPE of 0.4%, demonstrating its strength in capturing short-term price fluctuations. The integration of seasonal indicators further improved predictive performance compared to univariate approaches, offering practical value for supporting timely policy interventions. This study is limited by its narrow feature set and the absence of external drivers such as weather conditions, production shocks, and distribution disruptions. Future research may incorporate additional exogenous variables or explore deep learning and hybrid ensemble methods to enhance robustness and generalizability.

Digitization and Virtualization of Minahasan Bamboo Instruments: Development of a Culturally-Informed Virtual Studio Technology Plugin

2025-10-09T05:08:07+00:00

Traditional musical instruments are an important component of intangible cultural heritage, yet many remain underrepresented in contemporary digital music production. In particular, Minahasan bamboo instruments face limited accessibility due to the lack of digital instrument representations. This study addresses this issue by developing a culturally informed Virtual Studio Technology (VST3) plugin that digitizes and simulates the characteristic sounds of Minahasan bamboo music. The proposed approach combines field recording, digital audio processing, and sample-based virtual instrument development using the JUCE framework. The resulting plugin was evaluated through functional, compatibility, performance, and cultural fidelity testing across multiple digital audio workstations. Experimental results demonstrate accurate MIDI-to-audio translation, low-latency performance, stable operation, and efficient CPU usage. User evaluations further confirm the authenticity of the sound and the usability of the interface. The findings indicate that VST-based digitization offers a practical and transferable solution for preserving traditional musical instruments while enabling their integration into modern music production workflows.